LLMs are trained on human data, so where does the tendency to add emojis come from? For example, when some models generate code explanations or even normal responses, they often add lots of emojis that people don’t really use that way in real life. My current guess (without having researched this ye
